Hi Kai, On Sat, Oct 10, 2009 at 11:22:05AM +0100, Kai Krueger wrote: > I can see that the errors by the backend servers are not haproxy's > errors or fault, but they are still errors to the end user and as a HA > solution they should try to be minimised. So it would be useful to > easily monitor these in haproxy. We have patched haproxy ( > http://api1.osm.absolight.net/haproxy?stats ) to show backend errors > in the 500 range as a separate column in the stats page.
I have nothing against adding such a thing in the stats. However, some users are already complaining that the stats page is large. I think we should compact some server errors into one column, and get the details by passing the mouse over the number. > On top of > that we have also patched it so that on 500 backend errors haproxy > automatically and immediately marks that backend server down until the > health checks confirm it is up again. Of cause, the latter is not safe > in general, the HTTP proxy scenario you give is a good example, but it > is nice to have it as an option and at least in our scenario it has > been very useful. But that is a different issue. What I want to do for this issue is to be able to define how many points of life are removed for how many consecutive errors. That way, it will be possible to say that if you observe, say 3 consecutive 5xx in a row, the server loses 1 health point (and switches to fastinter) or even more and immediately goes down. That's the only way I find to satisfy all users'needs which are all very different regarding this specific point. Regards, Willy
